Parallel Iterative Solvers for Ill-Conditioned Problems with Reordering
暂无分享,去创建一个
1. Preconditioned Iterative Solvers with Multicoloring In the previous work [1], author developed an efficient parallel iterative solver for finite-element applications on the Earth Simulator (ES) [2] using multi-level hybrid parallel programming model with MPI and OpenMP. The method employs three-level hybrid parallel programming model for SMP cluster architectures, consisting of MPI, OpenMP and vectorization. Multicolor-based reordering methods have been applied to distributed data sets on each SMP node in order to achieve optimum performance of Krylov iterative solvers with ILU/IC type preconditioning (Fig.1). Developed method attained 3.8 TFLOPS with 176 SMP nodes of ES, corresponding to more than 33% of the peak performance (11.3 TFLOPS) [1].
[1] K. Nakajima. Parallel Iterative Solvers of GeoFEM with Selective Blocking Preconditioning for Nonlinear Contact Problems on the Earth Simulator , 2003, ACM/IEEE SC 2003 Conference (SC'03).
[2] Kengo Nakajima,et al. The Impact of Parallel Programming Models on the Linear Algebra Performance for Finite Element Simulations , 2006 .